Where the Magic Happens: Combining Data and Plant Science

Since this practice is fairly new, I notice that there are many misconceptions and assumptions about combining plant science with data science and the introduction of new technologies in the greenhouse. Below I share some of the main misconceptions (and the truths) I frequently come across.

1. Misconception: “My gut feeling is accurate. I don’t need to use new technologies and apply data science in my company
   
   Truth:  There’s no question that data and AI could be very helpful for a company: smart algorithms can effectively identify correlations and data technology has the ability to record historical data. However, you need a good combination of plant science and data science. This means that the growers’ expertise (green knowledge or the green thumbs) need to be taken into account. The predictions, suggestions and recommendations are merely data driven support. I like to use the metaphor of a pilot flying a plane. Much of a pilot’s work is currently done by a computer. However, the pilot is still essential for making the actual decisions and taking informed decisions. Even the most successful growers can use data driven insights, data, predictions from time to time. We see more often that different greenhouse companies apply data science to improve the cultivation strategy.

2. Misconception: “Generic plant and data science models are valid for all growers.”
   Truth: Every grower is different and their specific information and growing conditions should be taken into consideration. Each greenhouse location, type of the greenhouse (lit or non-lit) and growing media type, etc. is unique. This variety directly affects growth and yield. Personalized data modelling and working with the right program is essential. Have a look at our own data platform, e-Gro, which can help you analyze and personalize your data in real-time.

3. Misconception: “Once I start working with data science and AI, I will immediately see results.”
   
   Truth: There’s an assumption that you’re presented with a solution once the data has been fed into the model. As you can imagine, it’s not that simple. Although you can perform descriptive analytics immediately, predictive analytics (using data for forecasting) and prescriptive analytics (the use of data in order to make better decisions) take more time. You’ll typically need to collect data for at least one growing season and it takes a lot of data science knowledge, experience and skills. The data needs a lot of preparation, processing, selection of the existing features and adding new ones (feature engineering). It’s not the raw data, but instead the refined insights that are of high value. Here is also where big data comes into play and the importance of working with the right people. More on this in my next article on tips for success.

4. Misconception: “Plant models will work directly as data science models and generic data science models will work directly on plants.”
   
   Fact: The “real-life” behavior of plants might considerably differ from the theoretical plant physiology models, as there are many external and unexpected factors involved in growing, such as extreme climate conditions, price fluctuations, pests and diseases, labor issues, etc. What might be valid in a lab type environment will not work directly in real life (one size does not fit all). This also works the other way around. You cannot assume that well known generic data science models and recipes will directly work on plants. These predictive models need be customized according to the domain knowledge of plants and horticulture specific practices.

5. Misconception: “Just hiring a data scientist and a developer will do the job.”
   
   Truth: As mentioned, it is very important to work with the right people, whether outsourced or in-house. A team to develop a valuable data science product should contain suitable data scientists, developers, domain experts (multiple), product managers and user experience designers, in order to make sure you are able to develop good algorithms. I suggest finding an established platform that supplies you with a comprehensive team to work with. Next to this, a lot of good data is needed from the greenhouse to build impactful algorithms, please read my next article about how to get the best out of your greenhouse data.

Not quite sure where to start? My advice would be to first learn data science basics and consult the right people. This takes reading relevant books and articles, some hands-on data wrangling, and trial and error as well. I would recommend following as many data science, machine learning, AI, plant science and horticulture related groups. Grodan is also a great resource for information and inspiration. Please do not hesitate to contact me with your questions, comments, thoughts or feedback. Excited for more? Read further practical tips in my next blog.

Further reading

• E-Gro: Grodan’s software platform, which offers real-time insights and smart recommendations based on the root zone, climate, crop and harvest data.
• IoT in Agriculture: A Bonus for All your Farming Activities